Skip to content

Grok computegpu computegpu_get_started

How to: Grok computegpu computegpu_get_started

client:Grok transport:streamable-http tool:computegpu_get_started

Add computegpu to Grok

  1. 01

    Open .mcp.json (in your project root) and merge this in. Keep any servers already there.

{
  "mcpServers": {
    "computegpu": {
      "type": "http",
      "url": "https://mcp.computegpu.com/mcp"
    }
  }
}
  1. 02

    Save the file and restart Grok.

  2. 03

    Ask for something computegpu_get_started does. Grok lists the server's tools on connect and calls computegpu_get_started itself — you do not invoke it by name.

  3. 04

    If nothing happens, check the server is running and that computegpu's identifier in your config matches the one above exactly.

Grok Build reads Claude Code's .mcp.json and Claude Desktop's config unchanged, so a server you have already set up there needs nothing new. The Grok web app and X integration do not take custom servers. Grok docs

Set these first

computegpu will not start until these are set, so computegpu_get_started never becomes available.

  • COMPUTEGPU_API_KEY

Same tool, other clients

This list was read from the server itself, by connecting to it and calling tools/list on 16 October 2026. It is what the server actually exposes, not what its listing claims.

Mutating and Read-only are read off each tool's name, not its schema — a hint, not a guarantee. The registry stores tool names only; connect the server for its live schemas.